We first tested for the effect of covariates on maxdrinks. As expected, gender was a highly significant predictor of maxdrinks (P=0.0001). Because the COGA dataset consists of large multi-generation families we used birth cohort to account for the secular effects on maxdrinks across generations. We divided subjects into 4 cohorts based on their year of birth (< 1930< 1930 – 1949 – 1950 – 1969, and ≥1970). The effect of age at interview was not independent of birth cohort, therefore it was not used as a covariate in the final association analysis. Quantitative trait association analysis was performed using a mixed linear model (MLM) implemented in the GWAF (Chen and Yang 2010) package of R, using gender and birth cohort as covariates and natural log transformed maxdrinks as the trait. The correlated nature of the COGA sample was accounted for using random effects correlated according to the degree of relatedness within a family using a kinship matrix.
